Nvidia backs Indian startup helping companies run AI models with fewer GPUs

Nvidia joined a $14 million Series B investment in Simplismart, alongside Accel, Shastra VC, and Micromax. Dallas Venture Capital led the round. Simplismart matters because companies increasingly need to run AI models reliably without wasting expensive computing capacity. Its software helps manage the difficult trade-offs among speed, cost, and performance.

The platform sits between an AI model and the infrastructure running it. It adjusts deployment settings to match a customer’s needs. It can help models respond faster, handle more requests, or consume less computing power. The company also offers fine-tuning, automatic scaling, and performance monitoring across cloud and on-premises systems.

Simplismart says InVideo cut GPU serving costs by 56% using its software. Its customers include HeyGen, Higgsfield, Ema, and Yotta. Nvidia is also integrating Simplismart with its Dynamo inference software and helping connect it with potential customers and partners.

AI inference is the process of using a trained AI model to produce an output. That output might be a prediction, generated image, video, text response, or other result. Inference happens when a customer uses an AI application. It therefore affects response time, capacity, reliability, and operating cost.

AI models perform enormous numbers of mathematical operations. GPUs are designed to handle many similar operations in parallel. That makes them well suited to running models, especially when many requests arrive together. The article describes GPUs as expensive resources used to power AI applications and inference services.

Inference still requires careful management. Different workloads need different balances between speed, cost, and performance. Simplismart’s software adjusts how models run on available infrastructure. It can also scale resources as demand changes and monitor performance, helping companies use their GPU capacity more effectively.

Simplismart raised $14 million in an all-primary Series B funding round. Nvidia participated with Accel, Shastra VC, and Micromax, while Dallas Venture Capital led the investment. Because the round was all-primary, the funding went into the company rather than buying existing shares, as described in the article.

The investment valued the four-year-old startup at about $100 million, including the new money. Simplismart was founded in 2022 by Amritanshu Jain and Devansh Ghatak. It now employs 58 people, according to Jain. The company serves customers in both India and the United States.

Its annualized revenue run rate increased 14-fold over the previous six months. Jain described it as an eight-figure sum but did not disclose the exact amount. The figure reflects recent customer usage, not contracted recurring revenue, and excludes revenue passed through to infrastructure providers.

When a model serves more requests on the same GPU capacity, the company can spread its infrastructure cost across more outputs. If it can serve the same workload with fewer GPUs, it may reduce spending on computing resources. This matters because companies are investing heavily in GPUs, while AI inference must remain economical at scale.

Better utilization can also increase capacity and reduce waiting during busy periods. Simplismart says its software adjusts deployment to control response time, request volume, and computing consumption. It can automatically scale resources as demand changes. These settings help match infrastructure behavior to a specific application rather than applying one fixed configuration.

The result is not automatically “cheapest” or “fastest” inference. Companies choose trade-offs based on their service requirements. Simplismart cites InVideo as an example: the AI video platform cut GPU serving costs by 56% using the startup’s software. The article does not provide separate performance figures for that case.

Nvidia’s investment appears paradoxical because Simplismart can help customers run some workloads with fewer GPUs. Nvidia is the dominant supplier of AI chips, so fewer chips in one deployment might seem to threaten hardware demand. The article presents this tension directly.

Simplismart’s CEO, Amritanshu Jain, argues that more efficient inference can ultimately benefit Nvidia. If companies get better results from their infrastructure, AI applications may become more affordable and practical. That could encourage wider adoption and make Nvidia hardware more attractive to customers. Efficiency can therefore support the broader market, even when a particular workload needs fewer GPUs.

Nvidia is also working with Simplismart on technical integrations, including its Dynamo inference software. Nvidia’s solutions architects are collaborating with Simplismart engineers. The company has also helped introduce the startup to potential customers and partners. These activities give Nvidia a role in the growing inference-software ecosystem.

Simplismart competes with inference tools from Nvidia and major cloud providers, including Amazon Web Services, Google Cloud, and Microsoft Azure. The article places all of them in a crowded market shaped by rising AI demand and pressure to use costly GPU resources more effectively.

Its stated difference is customization. Simplismart says rival offerings often provide preset choices focused on faster or cheaper inference. Simplismart instead tries to find a balance for each customer. Its software adjusts deployment and management settings to meet specific targets for response speed, request handling, and total computing cost.

The startup says it tends to win when customers have strict service-level agreements and tight total cost-of-ownership requirements. It can run through its own platform or inside a customer’s cloud or on-premises infrastructure. It also supports fine-tuning, automatic scaling, monitoring, and cloud providers building their own inference services.

Modern AI models require large amounts of computation to process inputs and generate outputs. GPUs are important because they can perform many related mathematical operations in parallel. That makes them well suited to training and running AI models. The article describes them as the infrastructure companies are buying heavily to power AI applications.

The economic challenge is utilization. A company may own substantial GPU capacity but still struggle to run models reliably and economically at scale. Different workloads make different demands on response speed, request volume, and performance. Poorly matched configurations can waste capacity or fail to meet business requirements.

Better software can improve the value of existing hardware. Simplismart says it adjusts model deployment, scales resources as demand changes, and monitors performance. Its reported 56% reduction in InVideo’s GPU serving costs illustrates the potential. As AI applications grow, efficient inference could become increasingly important to controlling operating costs and expanding access.